See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/234037268

Jiles-Atherton Magnetic Hysteresis Parameters Identification

Article  in  Acta Physica Polonica Series a · September 2011

DOI: 10.12693/APhysPolA.120.491

CITATIONS READS

36 2,733

2 authors:

Nicusor Cristian Pop Ovidiu Florin Caltun

Universitatea Alexandru Ioan Cuza Universitatea Alexandru Ioan Cuza
6 PUBLICATIONS   66 CITATIONS    162 PUBLICATIONS   1,926 CITATIONS   

SEE PROFILE SEE PROFILE

Some of the authors of this publication are also working on these related projects:

This work is to understand the influence of divalent substitutions on the electromagnetic properties of Ni-Zn ferrites. View project

Special issue in Materials ( IF = 3,057) View project

All content following this page was uploaded by Ovidiu Florin Caltun on 22 January 2014.

The user has requested enhancement of the downloaded file.

Vol. 120 (2011) ACTA PHYSICA POLONICA A No. 3

Jiles–Atherton Magnetic Hysteresis Parameters Identification

N.C. Pop∗ and O.F. Caltun
“Al. I. Cuza” University, Faculty of Physics & Carpath Center, 700506 Iasi, Romania

(Received January 28, 2010; revised version December 3, 2010; in final form February 23, 2011)

This paper presents a method of fitting the magnetic hysteresis curves using a calculating algorithm within a
computer programme. The calculating programme uses the equations of the Jiles–Atherton model to simulate the
M (H) magnetization curve. The aim of this program is to find the values of the parameters which are part of the
Jiles–Atherton model once the regression curve for a major magnetization curve obtained experimentally is found.
The values of these parameters can be increased or decreased gradually. The parameters modification takes place
in such a way that the root mean square deviation between the points of the two graphs, also experimentally
simulated, should decrease gradually and, in the end, it should be found below the preestablished percentage value
in comparison to the previous value. The procedure for minimizing the root mean square deviation has as a result
the fact that the simulated curve represents the regression curve. The calculating programme considers a second
method of calculating the values of the parameters by using the already found regression curve. This paper also
presents the values of the Jiles–Atherton model parameters obtained by fitting the magnetization curves of several
cobalt ferrites samples, sintered at different temperatures.

PACS: 75.60.–d, 75.60.Ej

1. Introduction two ways. The algorithm made to obtain the regression

curve is based on repeated calculating structures which
This paper presents a calculating algorithm with two
gradually draw the simulated major magnetization curve
complementary methods of determining the regression
nearer to the experimental one. This approach is based
curve in the case of major experimental magnetization
on a simple principle: the a, α, k, c coefficients are grad-
curves. For this purpose the equations of the Jiles–
ually increased or decreased so that in the end the differ-
Atherton (J–A) model [1] are used, obtaining the follow-
ence between the graphs should be minimum. Within the
ing simulated magnetization curves: the Man (H) anhys-
calculating programme the initial values of these param-
teretic curve, the M (H) first magnetization curve with
eters are random. The root mean square deviation could
H increasing from 0 to +Hmax and the major magnetiza-
be considered as measure of the distance between the
tion curve with H in the [−Hmax , +Hmax ] interval. The
simulated curve and the experimental one and is given
major curve is given by two equations, as it will be seen
by the equation
in Sect. 2, according to the way H increases or decreases s
in the mentioned range. Hmax is the maximum value of PN 2
i=1 (Mi − Mexp,i )
the applied magnetic field which magnetizes a ferro or S= , (1)
N
ferrimagnetic sample.
In the J–A model equations there are four parameters: where N is the number of points on the experimental ma-
a, α, k and c. To obtain the regression curve it is nec- jor magnetization curve, Mi represents the value of mag-
essary to find the values of these parameters. Of course, netization calculated by the computer programme for a
attempts have been made to find the regression curve field identical or very close to the experimental Hexp,i and
for a given experimental magnetization curve and to cal- Mexp,i is the value of magnetization experimentally de-
culate the coefficients of the J–A model using different termined for Hexp,i . The calculating programme achieves
programming methods or different mathematical adjust- a leveled variation of the magnetic field. The calculated
ments of the used equations to simplify the achievement values of H do not exactly overlap the Hexp,i experimen-
of this objective [2–12]. Taking into consideration that tal values but some of them could be “as close as possi-
one of the model’s equations is a differential equation ble”. In other words, the terms of a subsequence of the
which does not have an analytical solution, the attempts H values are very close to the sequence of the Hexp,i ,
of fitting experimental magnetization curves were based i = 1 . . . N experimental values. Minimizing S leads to a
on numerical methods. Unlike the already taken actions hysteresis loop having approximately the same shape and
[5–8], what we suggest is a calculating algorithm which area as the experimental one, as it will be seen in Sect. 4.
determines the values of the J–A model parameters in The simulated curve obtained through minimizing the
root mean square deviation is called regression curve and
has the property that the sum of the squares of the dis-
tances, measured along the ordinate, from its points to
∗ corresponding author; e-mail: [email protected] the experimental ones is minimum. Thus the distance be-

(491)
492 N.C. Pop, O.F. Caltun

tween two points on the two curves, corresponding to the of the magnetization anhysteretic and the irreversible one
same magnetic field, is the difference of the magnetiza-
Mrev = c(Man − Mirr ). (4)
tions which appears in Eq. (1): Mi − Mexp,i . The values
of the J–A model coefficients for three cobalt ferrite sam- Finally, the value of the total magnetization is given by
ples obtained through sintering at different temperatures the following equation:
will also be shown in Sect. 4.
M = cMan + (1 − c)Mirr . (5)

2. The equations of the J–A model From (5) results that

dM dMan dHe dMirr
The magnetization calculated according to the J–A =c + (1 − c) (6)
dH dHe dH dH
model has two components, an irreversible and a re- and
versible one corresponding to the irreversible or reversible m − cMan
phenomena which take place within the magnetic mate- Mirr = . (7)
1−c
rial during the magnetization. The domain walls mov-
Substituting (7) into (2) we obtain
ing over the pinning sites represent irreversible displacing
and their movement between two successive pinning sites Man − M
dMirr = dH . (8)
or the rotations of the magnetic moments which lead to kδ(1 − c) − α(Man − M )
the decrease of the magnetization at field inverting after
achieving the saturation represent reversible phenomena. Differentiating (3) we get
" µ ¶2 µ ¶2 #
Thus the total magnetization is M = Mrev + Mirr . dMan Ms He a
The irreversible component of the magnetization is = 1 − coth + . (9)
dHe a a He
given by the following differential equation:
But
Man − Mirr dHe dM
dMirr = dH, , (2) =1+α . (10)
kδ − α(Man − Mirr ) dH dH
where Man is the anhysteretic magnetization and k is a Substituting (8), (9) and (10) into (6) we obtain the equa-
coefficient which appears as a factor of proportionality tion
between the wasted energy when a domain wall crosses a ½ " µ ¶2 µ ¶2 #
dM M s H e a
pinning site and the variation of the sample’s magnetiza- χ0 = = c 1 − coth +
tion as an effect of this movement. Within this equation dH a a He
also, δ = 1 if dH/ dt > 0 and δ = −1 if dH/dt < 0. ¾.
Man − M
The anhysteretic magnetization is the magnetiza- + (1 − c) (1 − αc) . (11)
tion of an ideal ferromagnet which does not have pin- kδ(1 − c) − α(Man − M )
ning sites and thus, its magnetization curve does not This is the final differential equation that can be reached
present hysteresis, the magnetization being made only using J–A model, equation which gives the dependence
through the rotating mechanism of the magnetic mo- between the magnetization of a magnetic sample and the
ments. Langevin’s function specific to magnetizing para- applied magnetic field.
magnetic substances is used within the J–A model for an-
hysteretic magnetization, in which case an effective field, Jiles et al. [2, 3] obtained the a, α, k, c coefficients
He , within the sample replaces the magnetic field, H, ex- which appear in Eqs. (2)–(5) or (11) by calculating
ternal to the sample. The equation of the anhysteretic some differential magnetic susceptibilities at different
magnetization is points of the anhysteretic magnetization curves, major
µ ¶ or first magnetization. These differential susceptibilities,
He a χ0 ( dM/ dH or dMan /dH) are the derivates of the mag-
Man = Ms coth − , (3)
a He netization in relation to the H field at different points
where a = kB T /(µ0 m) is another parameter within the of the magnetization curve or at different points of the
model which is obtained as a result of the fitting pro- anhysteretic curve. The initial, remanent, coercive and
cess, m is the magnetic moment of a typical magnetic saturation anhysteretic differential susceptibilities will be
domain and He =H+αM (the polycrystalline ferromag- marked by χ0an , χ0anmax , χ0anc , χ0anm . The differential sus-
net is treated [1] as a canonical ensemble of interacting ceptibilies on the major magnetization curve correspond-
magnetic domains each carrying a magnetic moment m). ing to the mentioned areas will also appear in these equa-
Sablick nad Jiles [13] consider that Ms = mn, where tions: χ0in , χ0max , χ0c , χ0m .
Ms is the saturation magnetization and n is the density The Taylor series expansion of the anhysteretic mag-
of magnetic domains of feromagnetic sample. Therefore netization in the vicinity of H = 0, M = 0, gives
n = µ0 Ms a/(kB T ) and n ∼ aMs . a is a coefficient which
He Ms
can be interpreted as a measure of the coupling between Man ∼= Ms = (H + αM ) . (12)
the adjoining magnetic domains. The reversible compo- 3a 3a
nent of the magnetization is a fraction c of the difference Substituting (5) into (12) we obtain
 Jiles–Atherton Magnetic Hysteresis Parameters Identification 493

Ms £ ¤ 2) Calculation of the coordinates of the simulated mag-

Man = H + αcMan + α(1 − c)Mirr . (13)
3a netization curve;
For H = 0 we will have Man = 0. We consider that in 3) Calculation of the average square deviation S using
this zone the irreversible component of the magnetization Eq. (1);
is null (for very small fields the domain walls perform re- 4) Gradual increasing or decreasing the a, α, k, c pa-
versible motions). Mirr = 0. What results from Eq. (2) rameters in order to minimize S;
is ¯ 5) Displaying the calculated physical units including
dMirr ¯¯ ∼ the values of the a, α, k, c parameters, of the experimen-
= 0. (14)
dH ¯ H=0
tal and regression magnetization curves;
6) Recalculation of the values of the a, α, k, c param-
Taking into consideration Eqs. (5) and (14) we can write eters using Eqs. (16)–(18), (20) and displaying of these
¯ ¯
0 dM ¯¯ ∼ dMan ¯¯ values.
χin = =c . (15)
dH ¯H=0 dH ¯H=0 Detailing stage 1:
The experimental data are “read” first so that a set of
As a matter of fact, Eq. (15) presents c as the relation be- variables corresponding to the magnetic field gets the cor-
tween the differential magnetic susceptibilities calculated responding numerical values from a text file. Another set
in the beginning zone of the first magnetization curve of variables receive the values corresponding to the mag-
χ0in netization. Using the second set of variables the value of
c∼
= . (16) the saturation magnetization Ms is determined through
χ0an
interpolation. The magnetic field is gradually increased
The following two equations are presented as they were or decreased with dH. If dHexp is the average distance
calculated by Jiles et al. [3]: between two successive values of the measured magnetic
à ! field intensity, in order to obtain calculated values of the
Man (Hc ) 1
k= α+ , (17) magnetic field intensity as close as possible to the exper-
1−c χ0max − c dMan (Hc )
1−c dH imental ones, it is necessary that dH should be much
smaller than dHexp . Thus the preestablished vicinity of
k
MR = Man (MR ) + . (18) a random experimental value of the Hexp,i magnetic field,
α 1
1−c + dMan (MR ) in which there is only one calculated value H of the mag-
χ0rem −c dH
netic field, will be V = (Hexp,i − dH/2, Hexp,i + dH/2).
MR is the remanent magnetization, Man (MR ) = Man (0) If the distance between any two consecutive experimental
and dH/ dt < 0, dMan (MR )/ dH= dMan (0)/ dH, values of the magnetic field is constant then dH= dHexp .
dH/dt < 0 [2, 3]. In the same stage the a, α, c, k parameters will receive
The c coefficient being known from (16), k and α, re- random values. The maximum value of the calculated
spectively, can be found from (17) and (18). Hmax is initialized with the maximum experimental value
Differentiating (13) in H = 0 and taking (14) into ac- of the magnetic field. Also, the initial values of the cal-
count, we have culated magnetization M and of the calculated magnetic
¯ µ ¯ ¶ field H are null.
dMan ¯¯ ∼ Ms dMan ¯¯
= 1 + αc . (19) Detailing stages 2 and 3:
dH ¯ H=0 3a dH ¯ H=0 The calculation of the coordinates (H, M ) on the sim-
Equation (19) leads to ulated magnetization curve takes place as follows:
— for a given H the effective He field is calculated
Ms αcMs
a∼
= 0
+ . (20) using the equation He =H+αM (initially H and M are
3χan 3 null);
In conclusion, if the graph of a major magnetization curve — the anhysteretic magnetization is calculated using
is known, including the anhysteretic and the first mag- Eq. (3);
netization curves, the a, α, k, c coefficients can be cal- — the variation of the irreversible component of dMirr
culated using Eqs. (16)–(18) and (20) first obtaining the magnetization is calculated using Eq. (2), where dH is
susceptibilities involved in these equations. the variation step of the calculated magnetic field;
— the value of the irreversible component of the mag-
0 0
netization is found: Mirr = Mirr + dMirr , where Mirr is
3. The method of fitting the experimental the previous value of this magnetization corresponding to
hysteresis curves the previous point on the simulated magnetization curve;
— finally, the total magnetization M , which includes
The structure of the calculating programme which fits the irreversible component is calculated. Equation (5) is
the experimental magnetization curve has the following used;
calculating steps: — after calculating M for a given H field,
1) Initialization of the physical units which are the ob- the value of the field becomes H + dH on the
ject of the calculation; 0 → Hmax , −Hmax → +H max intervals and H − dH on
494 N.C. Pop, O.F. Caltun

the +Hmax → −H max interval; after this modification of according to the same procedure as the increasing mod-
the field the sequence of operations described above starts ification. Therefore, the gradual modifications of these
again. The repetition of the calculation sequence stops parameters lead to the step-by-step decrease of S. Af-
after the values were included once in the intervals and ter the successive modifications of all the parameters the
reached their extremes each time. The first interval cor- program evaluates if the difference between the last two
responds to the calculation of the coordinates of the sim- values of S is smaller than a certain percentage of S’s last
ulated first magnetization curve and the other two to- value. If this condition is not fulfilled, then the series of
gether correspond to the simulated major magnetization modifications of these parameters starts again. If this
curve. condition is fulfilled, then the z fraction decreases twice
The calculation of the average square deviation S takes and the succession of the parameters’ modifications re-
place simultaneously with the previously mentioned cal- sumes. The decrease of z implies a finer modification of
culations. each of the parameters thus a more precise calculation
The calculation of S takes place as follows: of them. The decrease of the z fraction takes place until
— at each repetition of the calculation sequence pre- z < 0.001%.
viously described, for each H value of the magnetic field Then the computer screen displays the experimental
calculated in the variation intervals +Hmax → −H max and the calculated magnetization curves, the a, α, k,
and −Hmax → +H max (the major magnetization curve), c coefficients, the remanent magnetization, the coercive
it is tested if this value is in the preestablished vicinity V field and the susceptibilities mentioned in Sect. 2.
of an experimental value Hexp,i , i = 1 . . . N , N being the The susceptibilities mentioned in Sect. 2 are the tan-
number of experimental data on the major experimental gents to the magnetization curve at different points, and
magnetization curve; the programme calculates them simultaneously with the
— if the previous condition is fulfilled, then the value coordinates of the simulated magnetization curve.
of the calculated magnetization M (H) is memorized by a Stage 6:
variable of the programme, Mi , and then the next modi- Immediately after finalizing the fitting, there is the
fication of the square of S takes place: S 2 → S 2 + [Mi − possibility for the user to calculate the a, α, k, c pa-
Mexp,i ]2 . If the condition is not fulfilled, S 2 keeps its rameters using Eqs. (16)–(18), (20). By pressing a but-
value. Mexp,i is the measured magnetization correspond- ton on the interface the values of the parameters calcu-
ing to the measured magnetic field Hexp,i . At first S = 0. lated in this way are displayed. These new values can be
When ending these calculations, after all the coordi- compared to those obtained through minimizing the root
nates of the simulated magnetization curve have been mean square deviation. Therefore, there are two com-
calculated, the value of S becomes (S 2 /N )1/2 ; thus the plementary methods of calculating the parameters of the
calculation of the average square deviation is finalized J–A model. The first method corresponds to the values
according to Eq. (1). directly obtained through fitting and the second method
Stage 4 follows, which means the modification of one of corresponds to the use of the above mentioned equations
the four parameters. This modification aims at obtaining immediately after finding the regression curve. The sec-
a simulated curve closer to the experimental one, this ond method would not be possible without finding the
being possible only if, as a result of the modification, the regression curve because all the susceptibilities which are
value of S is smaller than the previous one. part of these equations are calculated using this curve.
Detailing stage 4:
After modifying one of the mentioned parameters with 4. Results
a fraction z of its previous value, for example increasing,
the calculation programme recalculates the coordinates The experimental data used by the application are de-
of the simulated curve M (H) and recalculates the aver- termined by a vibrating samples magnetometer (VSM).
age square deviation S. In this way, a repetition of stages We will further present the fitting results for three sam-
2–3 takes place. If S’s new value is smaller than the pre- ples of CoFe2 O4 obtained by coprecipitation method and
vious one, the modification of that parameter continues sintered for 5 h in air atmosphere at temperatures of
increasing it again with the z fraction of its new value 1050 ◦C, 1100 ◦C and 1150 ◦C. The experimental and cal-
(for example, at the beginning z = 50%). The repetition culated values are presented in Table I. The experimen-
of this operation takes place until the value of S becomes tal and calculated hysteresis loop (regression curve) for
bigger than the previous one. Then, that parameter, and sample 1 (sintered at 1050 ◦C) are presented in Fig. 1. In
also S, returns to the value considered before the last Table II there are presented the values of the J–A model’s
modification. After this the modification of the next pa- parameters for sample 1 obtained by the two calculation
rameter starts, etc. If from the very first increasing mod- modes. We can observe that in the case of these ma-
ification of one of the parameters, the decrease of S is not terials the values of the a, k, α parameters determined
possible, then there is an alternative option to have a de- with the two methods are relatively close. The param-
creasing modification. If this does not have as a result eters used to determine the quality of the fittings were
the decrease of S, then the next parameter is taken into the quality factor, ε [12], and the Pearson coefficient, r2 .
consideration. The decreasing modification takes place The expression of the quality factor is given by equation
 Jiles–Atherton Magnetic Hysteresis Parameters Identification 495

S — the increase of c with the sintering temperature in-

ε= . (21)
Ms dicates an augmentation of the reversible component of
It represents a percentage value of the root mean square the magnetization which can be caused by the decrease
deviation. of the blocking areas density.

TABLE II
A comparison between the values of the Jiles coeffi-
cients for sample 1 obtained through the two modes
of calculation.

a k
α c
[A/m] [A/m]
mode 1 33679 19083 0.178 1 × 10−6
mode 2 33661 18964 0.176 9 × 10−5

Considering the values for Ms and a from Table I, we

can say that in the studied temperature range, the den-
sity of the magnetic domains (n ∼ aMs ) of cobalt ferrite
Fig. 1. Regression curve for sample 1.
samples increases with the sintering temperature.

TABLE I 5. Conclusion
The values of the Jiles coefficients for the three samples
of CoFe2 O4 obtained through mode 2 of calculation. The designed software based on J–A model can be used
in fitting the experimental hysteresis loops of magnetic
Sample 1 Sample 2 Sample 3 materials with high accuracy. The purpose of the fit-
T [ ◦C ] 1050 1100 1150 ting programme was the well-precise determination of
ρ [g/cm3 ] 4.89 4.92 5.01 the values of the model’s coefficients for a certain ma-
terial. This purpose was achieved. Moreover, this calcu-
Ms [A/m] 414800 424000 439600
lating programme can also be used for fitting the mag-
Mrem [A/m] 174100 87600 77000 netization curves of samples under mechanical stress or
Hc [A/m] 18480 12610 7920 under other interaction conditions and, in this way, the
α 0.176 0.126 0.198 dynamic of the model’s coefficients in the given situations
k [A/m] 18964 12755 7978 can be observed. These coefficients, through their phys-
c 0.00009 0.00014 0.00023 ical meanings, can provide useful information regarding
the microstructure of magnetic materials and of the in-
a [A/m] 33661 36610 42860
teractions at this level.
r2 [%] 99.98 99.96 99.94
ε [%] 1 0.78 0.83
Acknowledgments

This work was supported by the European Social Fund

The r2 coefficient is given by the equation
PN in Romania, under the responsibility of the Managing
2
2 (Mi − Mexp,i ) Authority for the Sectoral Operational Programme for
r = 1 − PNi=1 ¡ ¢2 , (22) Human Resources Development 2007–2013 [grant POS-
i=1 M exp − Mexp,i DRU/88/1.5/S/47646].
where Mi is the calculated value of the magnetization for
the Hexp,i field and M exp is the arithmetic average of the References
magnetization experimental values. As the value of r2 is
closer to 1, the regression curve is more accurate. [1] D.C. Jiles, D.L. Atherton, J. Appl. Phys. 55, 2115
The values of these two quantities, presented in Table I, (1984).
show a relatively high precision of the obtained regres- [2] D.C. Jiles, J. Thoelke, IEEE Trans. Magn. 25, 3928
sion curves. Regarding the values obtained for the four (1989).
coefficients we can state that: [3] D.C. Jiles, J.B. Thoelke, M.K. Devine, IEEE Trans.
— the decrease of k with the increase of the sintering Magn. 28, 27 (1992).
temperature indicates a decrease of the blocking areas [4] D. Lederer, H. Igarashi, A. Kost, T. Honma, IEEE
density in the studied samples (in the considered tem- Trans. Magn. 35, 1211 (1999).
perature range) and thus a reduction of the energetic [5] P.R. Wilson, J.N. Ross, A.D. Brown, IEEE Trans.
losses; Magn. 37, 989 (2001).
496 N.C. Pop, O.F. Caltun

[6] F.R. Fulginei, A. Salvini, IEEE Trans. Magn. 41, [11] C. Cepisca, H. Andrei, V. Dogaru-Ulieru, J. Mater.
View publication stats
1100 (2005). Proc. Technol. 181, 172 (2007).
[7] J. Izydorczyk, IEEE Trans. Magn. 42, 3132 (2006). [12] J. Izydorczyk, J. Magn. Magn. Mater. 302, 517
[8] M. Toman, G. Stumberger, D. Dolinar, IEEE Trans. (2006).
Magn. 44, 1098 (2008). [13] M.J. Sablick, D.C. Jiles, IEEE Trans. Magn. 29, 2113
[9] H. Hauser, Y. Melikhov, D.C. Jiles, IEEE Trans. (1993).
Magn. 45, 1940 (2009).
[10] P. Vandewalle, J. Kovacevic, M. Vetterli, Signal Proc.
Mag., IEEE 26, 37 (2009).